This just in. Current Biology has published a short dispatch piece reviewing the flagellum evolution issue:
W. Ford Doolittle and Olga Zhaxybayeva (2007). “Reducible Complexity - The Case for Bacterial Flagella.” Current Biology, 17(13), R510-R512. July 3, 2007. DOI
I recently expressed some discouragement about the capabilities of blogs for critiquing scientific papers. I still have those reservations, but here is a data point that leans the other way:
Doolittle and Zhaxybayeva first summarize Pallen & Matzke (2006):
Last October, Pallen and Matzke  summarized in a review much of the relevant knowledge. Bacterial flagella are, in fact, diverse in composition (and quite distinct from archaeal analogs), but concerning eight axial bacterial proteins these authors inferred that “the flagellar rod-hook-filament complex has clearly evolved by multiple rounds of gene duplication and subsequent diversification, starting from just two proteins (a proto-flagellin and a proto-rod/hook protein)”. There are also many homology relationships to non-flagellar proteins. As Matzke puts it (personal communication): “of 42 so-called standard flagellar proteins in Escherichia coli/Salmonella, only 20 are universally required/detectable in all flagella, only 15 have no known homologs, and only two are both universally required and have no homologs.” So the stage seems set for a more detailed accounting of the events, intermediate functions and selective pressures giving rise to these amazing structures.
Then we get to Liu & Ochman (2007):
A giant step in that direction has now been claimed by Liu and Ochman . The paper quickly attracted (favorable) comment in ScienceNOW  and set off a firestorm of commentary (pro and con) in the evolutionary blogosphere. In ScienceNOW, Michael Lynch is quoted as saying “Complexity builds out of simplicity, and this [the Liu and Ochman paper] is a well-documented argument for how that can happen.” Maybe so, but there are some caveats we evolutionists should consider before hailing Liu and Ochman  as our next champions in the war against unreason. It is important that we scrutinize their arguments with special care, because they are likely to be under contention at the next trial.
This is a particularly important point – for better or worse, and as ridiculous as it seems outside of American battles over creationism, flagellum evolution has basically become a widely-followed exemplar for comparing (1) evolutionary reconstruction of the historical origins of complex structures, with (2) creationist claims that the natural origin of such structures is impossible. And it will almost certainly be featured if there is another creationism trial in the future – for example, the Discovery Institute’s new textbook, the one with the dastardly title Explore Evolution: The Arguments For and Against Neo-Darwinism (take a good look their website: make no mistake, this will be the future of creationism) doesn’t officially mention intelligent design. But it does devote a chapter to the bacterial flagellum and reiterating the usual half-baked ID arguments about it. (And yes, Explore Evolution copies the exact same Minnich-derived mistake that was pointed out in Pallen & Matzke (2006), and which the ID guys shamefully will not admit to.)
Hey, well at least the flagellum is different from the vertebrate eye, which has pretty much been done to death in evolution popularizations in my humble opinion.
Doolittle and Zhaxybayeva continue by reviewing the criticisms of Liu & Ochman’s proposal:
Liu and Ochman  present two conclusions about the evolutionary histories of the 24 ‘core’ flagellar genes in Bacteria that they consider ancestral for all flagellated bacteria. First, they assert that all 24 (not just eight) are homologous to each other, deriving from a single ancestor through successive duplications and diversifications, a sequence of events they reconstruct with phylogenetic analyses. Second, they argue that lateral gene transfer (LGT) has played only a minor role in the evolution of these 24 genes, that with only two exceptions “each of the genes has followed a common history in bacteria since they originated” (presumably at an early time, prior to the divergence of the major bacterial lineages).
The evidence presented for the first claim is the apparently significant BLAST scores between many individual flagellar genes, collectively uniting them all. This “single ancestor for all core flagellar proteins” hypothesis is, however, heavily criticized on the Panda’s Thumb weblog (http://www.pandasthumb.org/) by Matzke, [see note* - NM] who suggests that faulty setting of BLAST defaults has misled Liu and Ochman , and that homologies beyond those among axial proteins already noted are misinterpreted. Equally problematic, we think, is their conclusion that “proteins forming the flagellum, the rod, hook and filament proteins, originated in an order that mirrors the ‘inside-out’ flagellar assembly process”. Common sense might suggest such a scenario, but only rooted trees, which Liu and Ochman  do not provide, can prove it.
I guess it is safe to say that the eminent Ford Doolittle reads PT!
Doolittle and Zhaxybayeva then move on to another issue with the analysis in the Liu & Ochman paper, namely the paper’s conclusion that Lateral Gene Transfer (LGT) was relatively rare, effecting only four of the flagellar systems in the analysis. D&Z’s main point is that L&O’s trees of individual genes often have insufficient signal to statistically distinguish “LGT” from no “LGT”, and that the real conclusion in such situations should be “can’t tell with these data.” This issue is much closer to the cutting edge I think. It is a fairly common practice to stick together a number of genes to get more signal and higher-resolution phylogenies. I might be wrong, but I don’t think there is a standard test that is routinely applied to formally assess the statistical probability of LGT having occurred in each of the genes (if not, I might have a graduate research project right there). Presumably if LGT was a truly massive confounding issue, then stringing together genes would rarely or never produce phylogenies with improved resolution and statistical confidence, which I doubt many people would claim.
In the case of flagella, we have some background information that gives some reasons why many flagellar genes might be inherited together rather than being subject to a tremendous amount of individual LGT, e.g.: (1) the proteins form a coadapted complex and presumably the native protein is usually better adapted to function within the complex than a foreign replacement would be, (2) in the known cases of flagellar LGT (at least one of Liu & Ochman’s identified cases was known from earlier work), it is the whole flagellar system that gets laterally transferred (except for proteins like FliC – flagellin – which are highly variable and often exist in multiple divergent copies even in single genomes and single flagella, (3) LGT of both individual flagellar genes and whole flagellar complexes would presumably be easier between bacteria that are more closely related, because the flagellum has to interact with the inner membrane, cell wall, and outer membrane, and the latter two especially can vary dramatically between bacterial phyla, as can the cell’s system for regulating the construction of flagella, (4) if #3 is at least somewhat true, LGT will be both harder to detect and less relevant anyway for resolving the most interesting deep branches (most interesting if you are interested in flagellar origins, at least).
On the other hand, Doolittle is pretty keen on LGT – it’s his thing – and might not be otherwhelmed with general qualitative arguments. If, say, he was one of your reviewers on a paper, how would you convince him? One method might be to look at all of the flagellar genes under consideration and see if they are found near to each other in the bacterial genome. In some bacteria (particularly some convenient model organisms), many of the flagellar genes are grouped together in a unit, implying that they constitute a tightly-linked functional group, but often this is not the case. (Liu and Ochman were kind enough to send me some forthcoming work that bears on exactly this point; I won’t discuss it until published but it is clear it will be much less controversial than the PNAS paper.) It is also possible that one might measure generic LGT probability for particular genes by looking only at recent phylogenetic events where you have a high-resolution gene phylogeny, and use that to at least eliminate the genes that are highly subject to LGT. Anyhow, it’s a tough question and I’m not sure there will be an easy solution.
*Note: In all honesty this bit of it was worked out in discussions with Ian Musgrave and Doug Theobald, who deserve equal or more credit than me. I just blogged it, although I had already been beating the drums based on more general problems. And Liu & Ochman forthrightly volunteered the filters correction to correspondents.